Table 2 Comprehensive overview of CNN Layers used by the Proposed Hybrid Model.
From: Advanced air quality prediction using multimodal data and dynamic modeling techniques
Layer type | Input dimensions | Output dimensions | Activation | Filters/Units | Kernel size/stride | Purpose |
|---|---|---|---|---|---|---|
Input Layer | (128 × 128 × 3) | (128 × 128 × 3) | N/A | N/A | N/A | Accept raw satellite imagery data and Sensor Grid Data |
Conv2D Layer 1 | (128 × 128 × 3) | (128 × 128 × 32) | ReLU | 32 | (3 × 3/1) | Extract basic spatial features |
MaxPooling-2D Layer 1 | (128 × 128 × 32) | (64 × 64 × 32) | N/A | N/A | (2 × 2/2) | Downsample feature maps |
Conv-2D Layer 2 | (64 × 64 × 32) | (64 × 64 × 64) | ReLU | 64 | (3 × 3/1) | Extract deeper features |
MaxPooling-2D Layer 2 | (64 × 64 × 64) | (32 × 32 × 64) | N/A | N/A | (2 × 2/2) | Further, it reduces spatial dimensions |
Fully Connected 1 | (32 × 32 × 64) | 512 | ReLU | 512 | N/A | Learn high-level feature representation |
Fully Connected 2 | 512 | 256 | ReLU | 256 | N/A | Further abstraction for pollutant level |
Output Layer | 256 | 1 (or n classes) | Linear/SoftMax | 1/n | N/A | Predict AQI or pollutant concentrations |
